Bowden-PII

Local PII detection and redaction for Swiss and European identifiers.

Website · Downloads · Project guide · Blog · Roadmap · Contact

Version 0.0.1 is a developer preview. Detect supported identifiers on your own machine and replace them with typed placeholders before forwarding text.

This release provides a working Python API and command-line tool. The default rules-based engine has no runtime dependencies and does not need a model or network connection. It is early software for evaluation and integration work, not a validated general-purpose anonymization system.

The original-value map stays with your application. Send only the redacted text to the next service. The rules engine runs without a model or network connection.

Python 3.11 or newer is required. Clone the tagged release:

git clone --branch v0.0.1 --depth 1 https://github.com/48Nauts-Operator/bowden-pii.git
cd bowden-pii
python3 -m venv .venv
source .venv/bin/activate
python -m pip install .

On Windows, activate with .venv\Scripts\Activate.ps1 in PowerShell. You can also download the wheel from the release page and install it with python -m pip install bowden_pii-0.0.1-py3-none-any.whl.

bowden-pii "Contact mia@example.ch. AHV 756.9217.0769.85."
# Contact [EMAIL_1]. AHV [AHV_1].

The command reads standard input when no text argument is supplied:

printf '%s' 'Contact mia@example.ch' | bowden-pii --policy strict

Python API:

from bowden_pii import redact

result = redact("Contact mia@example.ch", policy="strict")
print(result.redacted)  # Contact [EMAIL_1]

detect(text) returns detected spans. redact(text) also returns public span metadata, an audit summary, and placeholder_map, which contains the original values. Keep that map local. The CLI's --json output includes it; do not treat that full JSON result as redacted output. Use result.redacted when forwarding text.

• Swiss AHV/AVS/NSS, UID/CHE, and VAT suffixes (MWST, TVA, IVA, TPV).
• IBAN, Swiss QR-IBAN, and credit cards.
• Email addresses, conservative Swiss phone patterns, URLs, IP addresses, and MAC addresses.

Where applicable, rules validate checksums and formats. A valid checksum does not establish that an identifier belongs to a real person or account.

Policies:

• strict: includes all supported deterministic identifier classes.
• balanced: preserves URLs.
• permissive: also preserves IP addresses, MAC addresses, and standalone UID values.

Repeated normalized values receive consistent placeholders within one result. Cross-document/session mapping is the integrating application's responsibility.

The source includes hybrid_detect, hybrid_redact, and the MiniLM adapter to preserve the experimental integration surface. No trained model weights, training datasets, or training tools are included in this release. A source checkout does not provide ready-to-run AI detection of names or addresses.

Advanced users may supply their own compatible detector to the hybrid API. The optional --mode hybrid --model-dir ... path additionally requires PyTorch, Transformers, and a compatible local token-classification checkpoint. It is not part of the out-of-the-box quickstart. The adapter currently processes at most 256 tokens by default and does not chunk long documents.

Detection can miss sensitive information or redact harmless values. Rules do not identify arbitrary names, addresses, or contextual personal details. Broader European coverage, long-text neural inference, independently labelled real-document evaluation, and model distribution remain development work. Synthetic fixtures and unit tests are useful checks, not proof of real-world privacy guarantees. Review results for your use case before relying on them.

This public snapshot contains the runtime, focused tests, and one small synthetic smoke fixture. Research collection/review tools, private history, large corpora, and model binaries remain outside this release.

python -m pip install -e ".[dev]"
pytest
ruff check .
ruff format --check .
bowden-pii-benchmark data/benchmarks/deterministic_smoke.jsonl

The benchmark compares identifier label counts, not exact character boundaries. GitHub Actions tests Python 3.11 and 3.14 and verifies that the package builds.

Email hello@bowden-pii.com for project enquiries. Report reproducible problems in GitHub Issues. Use synthetic examples in bug reports and leave out personal or confidential data.

See the visual roadmap for the next work and release notes for what each download includes.

Named in honor of Caspar Bowden. Independent and unaffiliated with his estate or any organization. See the website for background and sources.

MIT license. Runtime and focused tests were exported from Forgejo implementation commit ebeed975920048e42bbb44c89a4342e3dc528214; packaging and public documentation were prepared for this release. The internal Git history is not published here.